摘要
以使用嵌入式操作系统 Pocket Pc的个人数字助理(PDA)为实验平台研究了基于非特定人语音命令识 别的可定制声控拨号器。针对PDA存储空间和运算能力的限制,在保证性能的前提下从严格控制搜索空间和提高 解码速度出发,提出了结合搜索路径分数差值实时调整剪枝宽度的动态调整直方图剪枝策略,提出了利用速查表 加速似然计算的方法,并在通过实验验证后采用较少维数的特征、结合扩展声韵母进行声学建模等措施,有效地解 决了上述问题。在实际PDA设备上实验表明,在词表大小为200个人名时,识别正确率达98．70％,而识别速度比采 用标准算法的参考系统提高了约80倍,同时节省了约30％搜索存储空间。
Memory and speed are two demanding factors thai： must be faced when applying voice dialing speech recognition system to Pocket PC. We proposed a novel decoding method, dynamic histogram pruning adjusted by the difference scores of token paths,to precisely control the decoding search space and improve decoding efficiency. Besides,a new acoustic modeling method based on Extended Initial/Final（XIF） with less feature dimension is proven suitable for embedded speech recognition. By using the above methods developed,we implemented a speaker-independent and user definable voice dialing speech recognition system with good performance on a real PDA device. In 200-word-sized vocabulary,it obtained the accuracy of 98.70% and better recognition speed with 30% decoding space saving in comparison to the baseline system.
出处
《电声技术》
2005年第12期38-43,共6页
Audio Engineering
